open source

[ egocentric dataset ]

We believe our company will be measured by its net impact and research contribution.

[ about ]

Egocentric is the largest open source dataset of physical jobs. We are actively increasing the dataset size and quality.

[ LICENSE ]

Apache 2.0 license. Please credit us if you use our data.

[ v0.3 CHANGES ]

+2x action labels to 1 million labels
+2x dataset size to 0.1 years
+2x clip count to 6k clips
+2x size to 1.5 TB
Improved QA pipeline to filter more low-quality data.
76% of frames with 2 hands in view (95% CI, +-1%)

[DATA FORMAT]

Factories -> Workers -> Clips (mp4)/Metadata (JSON)

Each clip includes the following metadata: scenario, task_purpose and process_description; objects[] and skills[] arrays; and actions[] with frame-level timestamps.

[DOwnload INSTRUCTIONS]

1) Fill out the form below.
2) Open terminal and run bash download.sh